Themes for this doc

Basics

Reading Data

library(here)
movies <- read.csv(here("datasets", "movie_metadata.csv"))
gmsc_train <- read.csv(here("datasets", "gmsc_cs-training.csv"))

Setup Test/Train

train_idx <- sample(1:nrow(movies), size = floor(0.75 * nrow(movies)))
movies_train <- movies %>% slice(train_idx)
movies_test <- movies %>% slice(-train_idx)

Where 0.75 is the percentage (75%) of the data to put in the Training set.

Containers

# sequences
seq(1, 20)

 [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20

mySeq <- seq(1, 20, by = 3)
str(mySeq)

 num [1:7] 1 4 7 10 13 16 19

# vector
myVec <- c(1, 2, 3, 4, 100, 58, 5568)
myVec

[1]    1    2    3    4  100   58 5568

str(myVec)

 num [1:7] 1 2 3 4 100 ...

# character
myChar <- "hey!"
myChar

[1] "hey!"

vector_char <- c("jon", "Peter", "Sam")
vector_char

[1] "jon"   "Peter" "Sam"

str(vector_char)

 chr [1:3] "jon" "Peter" "Sam"

my_vec2 <- c(1, 3, 4, "Sam")

Factors

birth_month <- c("Dec", "May", "Apr", "Dec")
str(birth_month)

 chr [1:4] "Dec" "May" "Apr" "Dec"

birth_month <- factor(birth_month)
str(birth_month)

 Factor w/ 3 levels "Apr","Dec","May": 2 3 1 2

table(birth_month)

birth_month
Apr Dec May 
  1   2   1

month_levels <- c("Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", 
    "Oct", "Nov", "Dec")
birth_month <- factor(birth_month, levels = month_levels)
str(birth_month)

 Factor w/ 12 levels "Jan","Feb","Mar",..: 12 5 4 12

unique(birth_month)

[1] Dec May Apr
Levels: Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

table(birth_month)

birth_month
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 
  0   0   0   1   1   0   0   0   0   0   0   2

dplyr (Data Manipulation)

Can view help by vignette("dplyr") and vignette("two-table") or check out the online docs dplyr is a part of tidyverse

Filter

Can Use | as or

library("tidyverse")
table(movies$language)


           Aboriginal     Arabic    Aramaic    Bosnian  Cantonese 
        12          2          5          1          1         11 
   Chinese      Czech     Danish       Dari      Dutch   Dzongkha 
         3          1          5          2          4          1 
   English   Filipino     French     German      Greek     Hebrew 
      4704          1         73         19          1          5 
     Hindi  Hungarian  Icelandic Indonesian    Italian   Japanese 
        28          1          2          2         11         18 
   Kannada     Kazakh     Korean   Mandarin       Maya  Mongolian 
         1          1          8         26          1          1 
      None  Norwegian    Panjabi    Persian     Polish Portuguese 
         2          4          1          4          4          8 
  Romanian    Russian  Slovenian    Spanish    Swahili    Swedish 
         2         11          1         40          1          5 
     Tamil     Telugu       Thai       Urdu Vietnamese       Zulu 
         1          1          3          1          1          2

movies_language <- movies %>% filter(language == "Arabic")
movies_language

  color  director_name num_critic_for_reviews duration
1 Color    Serdar Akar                     16      122
2 Color Jehane Noujaim                     68      108
  director_facebook_likes actor_3_facebook_likes   actor_2_name
1                      11                    173 Bergüzar Korel
2                      63                      5   Ahmed Hassan
  actor_1_facebook_likes gross                         genres
1                    205    NA               Action|Adventure
2                    161    NA Documentary|Drama|History|News
    actor_1_name                 movie_title num_voted_users
1  Necati Sasmaz Valley of the Wolves: Iraq            14486
2 Khalid Abdalla                 The Square             6678
  cast_total_facebook_likes    actor_3_name facenumber_in_poster
1                       808 Ghassan Massoud                    3
2                       176   Aida Elkashef                    0
                             plot_keywords
1 abu ghraib|bomb|christian|explosion|iraq
2                                         
                                       movie_imdb_link
1 http://www.imdb.com/title/tt0493264/?ref_=fn_tt_tt_1
2 http://www.imdb.com/title/tt2486682/?ref_=fn_tt_tt_1
  num_user_for_reviews language country content_rating  budget title_year
1                  159   Arabic  Turkey                8300000       2006
2                   42   Arabic   Egypt      Not Rated 1500000       2013
  actor_2_facebook_likes imdb_score aspect_ratio movie_facebook_likes
1                    197        6.0         1.85                  467
2                     10        8.1         1.85                    0
 [ reached 'max' / getOption("max.print") -- omitted 3 rows ]

moviesSub <- movies %>% filter(language == "English" | language == "Spanish")
moviesBig_Spanish <- movies %>% filter(language == "Spanish" | budget > 10000000000)

movies_eng_sp <- movies %>% filter(language == "English" | language == "Spanish") %>% 
    filter(budget > 100000000)
dim(movies_eng_sp)

[1] 308  28

Arange

Order the rows

# single column
moviesBig_Spanish <- moviesBig_Spanish %>% arrange(desc(budget))

# arrange via multipe columns
moviesBig_Spanish <- moviesBig_Spanish %>% arrange(desc(budget), desc(title_year))

slice

Only see certain rows

movies_eng_sp %>% slice(c(35, 70))

  color    director_name num_critic_for_reviews duration
1 Color      Dan Scanlon                    376      104
2 Color Barry Sonnenfeld                     85      106
  director_facebook_likes actor_3_facebook_likes actor_2_name
1                      37                    760 Tyler Labine
2                     188                    582  Salma Hayek
  actor_1_facebook_likes     gross
1                  12000 268488329
2                  10000 113745408
                                     genres  actor_1_name
1 Adventure|Animation|Comedy|Family|Fantasy Steve Buscemi
2              Action|Comedy|Sci-Fi|Western    Will Smith
           movie_title num_voted_users cast_total_facebook_likes
1 Monsters University           235025                     14863
2      Wild Wild West           129601                     15870
  actor_3_name facenumber_in_poster
1   Sean Hayes                    0
2     Bai Ling                    2
                                            plot_keywords
1 cheating|fraternity|monster|singing in a car|university
2             buddy movie|general|inventor|steampunk|utah
                                       movie_imdb_link
1 http://www.imdb.com/title/tt1453405/?ref_=fn_tt_tt_1
2 http://www.imdb.com/title/tt0120891/?ref_=fn_tt_tt_1
  num_user_for_reviews language country content_rating    budget
1                  265  English     USA              G 200000000
2                  648  English     USA          PG-13 170000000
  title_year actor_2_facebook_likes imdb_score aspect_ratio
1       2013                    779        7.3         1.85
2       1999                   4000        4.8         1.85
  movie_facebook_likes
1                44000
2                    0

moviesBig_Spanish %>% slice(1:10)

  color director_name num_critic_for_reviews duration
1 Color  Joon-ho Bong                    363      110
2 Color  Carlos Saura                     35      115
  director_facebook_likes actor_3_facebook_likes       actor_2_name
1                     584                     74       Kang-ho Song
2                      98                      4 Juan Luis Galiardo
  actor_1_facebook_likes   gross                     genres actor_1_name
1                    629 2201412 Comedy|Drama|Horror|Sci-Fi    Doona Bae
2                    341 1687311              Drama|Musical  Mía Maestro
  movie_title num_voted_users cast_total_facebook_likes      actor_3_name
1   The Host            68883                      1173        Ah-sung Ko
2      Tango             2412                       371 Miguel Ángel Solá
  facenumber_in_poster                                 plot_keywords
1                    0        daughter|han river|monster|river|seoul
2                    3 dancer|director|love|musical filmmaking|tango
                                       movie_imdb_link
1 http://www.imdb.com/title/tt0468492/?ref_=fn_tt_tt_1
2 http://www.imdb.com/title/tt0120274/?ref_=fn_tt_tt_1
  num_user_for_reviews language     country content_rating      budget
1                  279   Korean South Korea              R 12215500000
2                   40  Spanish       Spain          PG-13   700000000
  title_year actor_2_facebook_likes imdb_score aspect_ratio
1       2006                    398        7.0         1.85
2       1998                     26        7.2         2.00
  movie_facebook_likes
1                 7000
2                  539
 [ reached 'max' / getOption("max.print") -- omitted 8 rows ]

select

Used to pick out certain variables

movies_sub <- movies %>% select(movie_title, director_name, gross, budget, )
dim(movies_sub)

[1] 5043    4

names(movies_sub)

[1] "movie_title"   "director_name" "gross"         "budget"

movies <- movies %>% select(movie_title, director_name, gross, budget, everything())
names(movies)

 [1] "movie_title"               "director_name"            
 [3] "gross"                     "budget"                   
 [5] "color"                     "num_critic_for_reviews"   
 [7] "duration"                  "director_facebook_likes"  
 [9] "actor_3_facebook_likes"    "actor_2_name"             
[11] "actor_1_facebook_likes"    "genres"                   
[13] "actor_1_name"              "num_voted_users"          
[15] "cast_total_facebook_likes" "actor_3_name"             
[17] "facenumber_in_poster"      "plot_keywords"            
[19] "movie_imdb_link"           "num_user_for_reviews"     
[21] "language"                  "country"                  
[23] "content_rating"            "title_year"               
[25] "actor_2_facebook_likes"    "imdb_score"               
[27] "aspect_ratio"              "movie_facebook_likes"

# using select to programmatically select several variables
moviesActors <- movies %>% select(starts_with("actor"))

Select Helpers

starts_with(): Starts with a prefix.
ends_with(): Ends with a suffix.
contains(): Contains a literal string.
matches(): Matches a regular expression.
num_range(): Matches a numerical range like x01, x02, x03.
one_of(): Matches variable names in a character vector.
everything(): Matches all variables.
last_col(): Select last variable, possibly with an offset.

rename

movies <- movies %>% rename(director = director_name)
names(movies)

 [1] "movie_title"               "director"                 
 [3] "gross"                     "budget"                   
 [5] "color"                     "num_critic_for_reviews"   
 [7] "duration"                  "director_facebook_likes"  
 [9] "actor_3_facebook_likes"    "actor_2_name"             
[11] "actor_1_facebook_likes"    "genres"                   
[13] "actor_1_name"              "num_voted_users"          
[15] "cast_total_facebook_likes" "actor_3_name"             
[17] "facenumber_in_poster"      "plot_keywords"            
[19] "movie_imdb_link"           "num_user_for_reviews"     
[21] "language"                  "country"                  
[23] "content_rating"            "title_year"               
[25] "actor_2_facebook_likes"    "imdb_score"               
[27] "aspect_ratio"              "movie_facebook_likes"

mutate

movies <- movies %>% mutate(budgetM = budget/1000000, grossM = gross/1000000, 
    profit = grossM - budgetM) %>% filter(budgetM < 400)

group_by

Director_Tot <- movies %>% group_by(director) %>% summarize(grossTotDir = sum(grossM, 
    na.rm = TRUE))
head(Director_Tot)

# A tibble: 6 x 2
  director        grossTotDir
  <fct>                 <dbl>
1 ""                    0    
2 A. Raven Cruz         0    
3 Aaron Schneider       9.18 
4 Aaron Seltzer        48.5  
5 Abel Ferrara          1.23 
6 Adam Carolla          0.106

Summerize

Director_Stats <- movies %>% group_by(director) %>% summarize(n = n(), min = min(grossM), 
    max = max(grossM), avg = mean(grossM), sd = sd(grossM))

Director_Stats %>% arrange(desc(max)) %>% slice(1:10)

# A tibble: 10 x 6
   director              n    min   max   avg    sd
   <fct>             <int>  <dbl> <dbl> <dbl> <dbl>
 1 James Cameron         7  38.4   761.  278.  301.
 2 Colin Trevorrow       2   4.01  652.  328.  458.
 3 Joss Whedon           4  25.3   623.  433.  282.
 4 Christopher Nolan     8  25.5   533.  227.  187.
 5 George Lucas          5 115     475.  348.  146.
 6 Andrew Adamson        4 142.    436.  284.  121.
 7 Francis Lawrence      5  58.7   425.  272.  135.
 8 Gore Verbinski        7  12.5   423.  190.  154.
 9 Roger Allers          2  84.3   423.  254.  239.
10 Lee Unkrich           1 415.    415.  415.   NA

Pulling it together

movies <- movies %>% mutate(genre_main = unlist(map(strsplit(as.character(movies$genres), 
    "\\|"), 1)), grossM = gross/1000000, budgetM = budget/1000000, profitM = grossM - 
    budgetM, ROI = profitM/budgetM)
movies <- movies %>% mutate(genre_main = factor(genre_main) %>% fct_drop())

Director_Avg <- movies %>% group_by(director) %>% summarize(num_movies = n(), 
    grossAvgDir = mean(grossM, na.rm = TRUE), profitAvgDir = mean(profit, na.rm = TRUE))

Director_Avg %>% arrange(desc(profitAvgDir)) %>% filter(num_movies > 1) %>% 
    slice(1:20)

# A tibble: 20 x 4
   director          num_movies grossAvgDir profitAvgDir
   <fct>                  <int>       <dbl>        <dbl>
 1 George Lucas               5        348.         277.
 2 Colin Trevorrow            2        328.         253.
 3 Joss Whedon                4        433.         250.
 4 Pierre Coffin              2        310.         237.
 5 Roger Allers               2        254.         189.
 6 James Cameron              7        278.         171.
 7 Pete Docter                3        313.         158.
 8 Francis Lawrence           5        272.         151.
 9 Irvin Kershner             2        173.         146.
10 Andrew Adamson             4        284.         131.
11 Joel Zwick                 2        145.         129.
12 George Roy Hill            2        131.         125.
13 Phil Lord                  4        178.         115.
14 Gary Ross                  3        183.         111.
15 Jon Favreau                7        223.         110.
16 Oren Peli                  2        108.         108.
17 Victor Fleming             3        110.         107.
18 Robert Wise                3        123.         101.
19 Christopher Nolan          8        227.         101.
20 Leonard Nimoy              3        118.         100.

Director_Avg %>% arrange(desc(num_movies)) %>% slice(1:20)

# A tibble: 20 x 4
   director          num_movies grossAvgDir profitAvgDir
   <fct>                  <int>       <dbl>        <dbl>
 1 Steven Spielberg          26       165.      99.5    
 2 Woody Allen               22        16.2     -0.00813
 3 Clint Eastwood            20        72.5     32.1    
 4 Martin Scorsese           18        57.2      0.0194 
 5 Ridley Scott              17        78.7     -5.43   
 6 Spike Lee                 16        21.9      5.26   
 7 Steven Soderbergh         16        65.7     25.5    
 8 Tim Burton                16       129.      51.5    
 9 Renny Harlin              15        34.9    -11.2    
10 Oliver Stone              14        52.3      7.67   
11 Barry Levinson            13        48.0     15.0    
12 John Carpenter            13        24.3      9.57   
13 Michael Bay               13       172.      49.6    
14 Robert Rodriguez          13        45.4     16.9    
15 Robert Zemeckis           13       125.      42.3    
16 Ron Howard                13       105.      28.5    
17 Sam Raimi                 13       171.      51.6    
18 Brian De Palma            12        50.1     10.3    
19 Joel Schumacher           12        63.2     13.8    
20 Kevin Smith               12        20.6      6.31

Director_Tot <- movies %>% group_by(director) %>% summarize(grossTotDir = sum(grossM, 
    na.rm = TRUE))
genre_collapse <- movies %>% group_by(genre_main) %>% summarize(Avg_ROI_genre = mean(ROI, 
    na.rm = TRUE), SD_ROI_genre = sd(ROI, na.rm = TRUE), SE_ROI_genre = sd(ROI, 
    na.rm = TRUE)/sqrt(n()), num_films = n())

actor_sum <- movies %>% group_by(actor_1_name) %>% summarize(Avg_ROI_actor = mean(ROI, 
    na.rm = TRUE), SD_ROI_actor = sd(ROI, na.rm = TRUE), SE_ROI_actor = sd(ROI, 
    na.rm = TRUE)/sqrt(n()), num_films = n())

actor_sum %>% filter(num_films > 2) %>% top_n(5, wt = Avg_ROI_actor)

# A tibble: 5 x 5
  actor_1_name     Avg_ROI_actor SD_ROI_actor SE_ROI_actor num_films
  <fct>                    <dbl>        <dbl>        <dbl>     <int>
1 Gunnar Hansen            246.         211.        122.           3
2 Jamie Lee Curtis          47.0         70.2        20.3         12
3 Jon Heder                 36.7         63.8        36.8          3
4 Madeline Kahn             37.4         10.7         4.79         5
5 Michael Emerson           22.3         32.1        18.5          3

genre_collapse %>% select(SE_ROI_genre)

# A tibble: 19 x 1
   SE_ROI_genre
          <dbl>
 1        0.291
 2        0.276
 3        2.07 
 4       12.3  
 5        0.470
 6        0.232
 7        1.67 
 8        0.296
 9        7.14 
10        1.11 
11       NA    
12       NA    
13       40.1  
14       11.4  
15        0.476
16        1.08 
17        0.618
18        1.44 
19        0.471

genre_collapse$SE_ROI_genre

 [1]  0.2909999  0.2755208  2.0667876 12.3284782  0.4697230  0.2315017
 [7]  1.6732844  0.2963527  7.1413626  1.1121905         NA         NA
[13] 40.1424230 11.4088478  0.4763281  1.0834822  0.6181190  1.4418917
[19]  0.4705548

magrittr (pipe opperator)

Can view help by vignette(“magrittr”) or check out the online docs magrittr is a part of tidyverse

We start with a value, here mtcars (a data.frame). Based on this, we first extract a subset, then we aggregate the information based on the number of cylinders, and then we transform the dataset by adding a variable for kilometers per liter as supplement to miles per gallon. Finally we print the result before assigning it. Note how the code is arranged in the logical order of how you think about the task: data->transform->aggregate, which is also the same order as the code will execute. It’s like a recipe – easy to read, easy to follow!

library(magritter)

car_data <- mtcars %>% subset(hp > 100) %>% aggregate(. ~ cyl, data = ., FUN = . %>% 
    mean %>% round(2)) %>% transform(kpl = mpg %>% multiply_by(0.4251)) %>% 
    print

  cyl   mpg   disp     hp drat   wt  qsec   vs   am gear carb       kpl
1   4 25.90 108.05 111.00 3.94 2.15 17.75 1.00 1.00 4.50 2.00 11.010090
2   6 19.74 183.31 122.29 3.59 3.12 17.98 0.57 0.43 3.86 3.43  8.391474
3   8 15.10 353.10 209.21 3.23 4.00 16.77 0.00 0.14 3.29 3.50  6.419010

Note also how “building” a function on the fly for use in aggregate is very simple in magrittr: rather than an actual value as left-hand side in pipeline, just use the placeholder. This is also very useful in R’s *apply family of functions.

The combined example shows a few neat features of the pipe (which it is not):

By default the left-hand side (LHS) will be piped in as the first argument of the function appearing on the
right-hand side (RHS). This is the case in the subset and transform expressions.
%>% may be used in a nested fashion, e.g. it may appear in expressions within arguments. This is used in the mpg to kpl conversion.
When the LHS is needed at a position other than the first, one can use the dot,‘.’, as placeholder. This is used in the aggregate expression.
The dot in e.g. a formula is not confused with a placeholder, which is utilized in the aggregate expression.
Whenever only one argument is needed, the LHS, then one can omit the empty parentheses. This is used in the call to print (which also returns its argument). Here, LHS %>% print(), or even LHS %>% print(.) would also work.
A pipeline with a dot (.) as LHS will create a unary function. This is used to define the aggregator function.

One feature, which was not utilized above is piping into anonymous functions, or lambdas. This is possible using standard function definitions, e.g.

car_data %>% (function(x) {
    if (nrow(x) > 2) 
        rbind(head(x, 1), tail(x, 1)) else x
})

  cyl  mpg   disp     hp drat   wt  qsec vs   am gear carb      kpl
1   4 25.9 108.05 111.00 3.94 2.15 17.75  1 1.00 4.50  2.0 11.01009
3   8 15.1 353.10 209.21 3.23 4.00 16.77  0 0.14 3.29  3.5  6.41901

However, magrittr also allows a short-hand notation:

car_data %>% {
    if (nrow(.) > 0) 
        rbind(head(., 1), tail(., 1)) else .
}

  cyl  mpg   disp     hp drat   wt  qsec vs   am gear carb      kpl
1   4 25.9 108.05 111.00 3.94 2.15 17.75  1 1.00 4.50  2.0 11.01009
3   8 15.1 353.10 209.21 3.23 4.00 16.77  0 0.14 3.29  3.5  6.41901

Additional Pipe Opperators

Tee %T>%

The “tee” operator, %T>% works like %>%, except it returns the left-hand side value, and not the result of the right-hand side operation. This is useful when a step in a pipeline is used for its side-effect (printing, plotting, logging, etc.). As an example:

rnorm(200) %>%
  matrix(ncol = 2) %T>%
  plot %>% # plot usually does not return anything. 
  colSums

[1]  6.464832 -6.553345

Exposition %$%

The “exposition” pipe operator, %$% exposes the names within the left-hand side object to the right-hand side expression. Essentially, it is a short-hand for using the with functions (and the same left-hand side objects are accepted). This operator is handy when functions do not themselves have a data argument, as for example lm and aggregate do. Here are a few examples as illustration:

iris %>% subset(Sepal.Length > mean(Sepal.Length)) %$% cor(Sepal.Length, Sepal.Width)

[1] 0.3361992

data.frame(z = rnorm(100)) %$% ts.plot(z)

Compound assignment %<>%

Finally, the compound assignment pipe operator %<>% can be used as the first pipe in a chain. The effect will be that the result of the pipeline is assigned to the left-hand side object, rather than returning the result as usual. It is essentially shorthand notation for expressions like foo <- foo %>% bar %>% baz, which boils down to foo %<>% bar %>% baz. Another example is

iris$Sepal.Length %<>% sqrt

The %<>% can be used whenever expr <- … makes sense, e.g.

x %<>% foo %>% bar
x[1:10] %<>% foo %>% bar
x$baz %<>% foo %>% bar

Plot

ggplot

ggplot2 is a system for declaratively creating graphics, based on The Grammar of Graphics. You provide the data, tell ggplot2 how to map variables to aesthetics, what graphical primitives to use, and it takes care of the details.

It’s hard to succinctly describe how ggplot2 works because it embodies a deep philosophy of visualisation. However, in most cases you start with ggplot(), supply a dataset and aesthetic mapping (with aes()). You then add on layers (like geom_point() or geom_histogram()), scales (like scale_colour_brewer()), faceting specifications (like facet_wrap()) and coordinate systems (like coord_flip()).

CHEATSHEET

Scatter

ggplot(mpg, aes(displ, hwy, colour = class)) + geom_point()

ggplot(data = movies) + geom_point(mapping = aes(x = imdb_score, y = budget))

Bar Plot

geom_bar makes the height of the bar proportional to the number of cases in each group (or if the weight aesthetic is supplied, the sum of the weights). If you want the heights of the bars to represent values in the data, use geom_col() instead. geom_bar() uses stat_count() by default: it counts the number of cases at each x position. geom_col() uses stat_identity(): it leaves the data as is.

ggplot(gmsc_train, aes(y = SeriousDlqin2yrs, x = age)) + geom_bar(stat = "identity", 
    fill = "red") + ggtitle("Serious Delinquencies in 2 years vs By Age") + 
    labs(x = "Age", y = "Serious Delinquencies in 2 years")

ggplot(gmsc_train, aes(y = SeriousDlqin2yrs, x = NumberRealEstateLoansOrLines)) + 
    geom_col(fill = "red") + ggtitle("Serious Delinquencies in 2 years vs Number of Loans or Lines") + 
    labs(x = "Number of Real Estate Loans Or Lines", y = "Serious Delinquencies in 2 years")

Jitter

The jitter geom is a convenient shortcut for geom_point(position = "jitter"). It adds a small amount of random variation to the location of each point, and is a useful way of handling overplotting caused by discreteness in smaller datasets.

ggplot(gmsc_train, aes(x = MonthlyIncome, y = SeriousDlqin2yrs)) + geom_jitter(alpha = 1/10, 
    color = "hotpink") + xlim(0, 100000) + ggtitle("Serious Delinquencies in 2 years vs Monthly Income") + 
    labs(x = "Monthly Income", y = "Serious Delinquencies in 2 years")

Histogram

Visualise the distribution of a single continuous variable by dividing the x axis into bins and counting the number of observations in each bin. Histograms (geom_histogram()) display the counts with bars; frequency polygons (geom_freqpoly()) display the counts with lines. Frequency polygons are more suitable when you want to compare the distribution across the levels of a categorical variable.

ggplot(diamonds, aes(carat)) + geom_histogram()

Box Plot

mpg %>% ggplot(aes(x = reorder(class, hwy), y = hwy, fill = class)) + geom_boxplot() + 
    xlab("class") + theme(legend.position = "none")

Line

In a line graph, observations are ordered by x value and connected.

The functions geom_line(), geom_step(), or geom_path() can be used.

x value (for x axis) can be :

date : for a time series data
texts
discrete numeric values
continuous numeric values

Data derived from ToothGrowth data sets are used. ToothGrowth describes the effect of Vitamin C on tooth growth in Guinea pigs.

len - Tooth Length
dose - Dose in mg (0.5, 1, 2)

See more here

df <- data.frame(dose = c("D0.5", "D1", "D2"), len = c(4.2, 10, 29.5))
head(df)

  dose  len
1 D0.5  4.2
2   D1 10.0
3   D2 29.5

# Basic line plot with points
ggplot(data = df, aes(x = dose, y = len, group = 1)) + geom_line() + geom_point()

# Change the line type
ggplot(data = df, aes(x = dose, y = len, group = 1)) + geom_line(linetype = "dashed") + 
    geom_point()

# Change the color
ggplot(data = df, aes(x = dose, y = len, group = 1)) + geom_line(color = "red") + 
    geom_point()

Observations can be also connected using the functions geom_step() or geom_path() :

ggplot(data = df, aes(x = dose, y = len, group = 1)) + geom_step() + geom_point()

ggplot(data = df, aes(x = dose, y = len, group = 1)) + geom_path() + geom_point()

Error Bars

ggplot(Genre_ROI, aes(y = ROI_avg, x = genre_main)) + geom_bar(stat = "identity") + 
    coord_flip() + ggtitle("Best Performing Genre By ROI") + geom_errorbar(mapping = aes(ymin = ROI_avg - 
    se_ROI, ymax = ROI_avg + se_ROI))

Themes

library("ggthemes") docs

Theme Library

Base example ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length,color = Species))

'data.frame':   150 obs. of  5 variables:
 $ Sepal.Length: num  5.1 4.9 4.7 4.6 5 5.4 4.6 5 4.4 4.9 ...
 $ Sepal.Width : num  3.5 3 3.2 3.1 3.6 3.9 3.4 3.4 2.9 3.1 ...
 $ Petal.Length: num  1.4 1.4 1.3 1.5 1.4 1.7 1.4 1.5 1.4 1.5 ...
 $ Petal.Width : num  0.2 0.2 0.2 0.2 0.2 0.4 0.3 0.2 0.2 0.1 ...
 $ Species     : Factor w/ 3 levels "setosa","versicolor",..: 1 1 1 1 1 1 1 1 1 1 ...

ggthemes

# For the complete list of themes in the package check
ls("package:ggthemes")[grepl("theme_", ls("package:ggthemes"))]

 [1] "theme_base"            "theme_calc"           
 [3] "theme_clean"           "theme_economist"      
 [5] "theme_economist_white" "theme_excel"          
 [7] "theme_excel_new"       "theme_few"            
 [9] "theme_fivethirtyeight" "theme_foundation"     
[11] "theme_gdocs"           "theme_hc"             
[13] "theme_igray"           "theme_map"            
[15] "theme_pander"          "theme_par"            
[17] "theme_solarized"       "theme_solarized_2"    
[19] "theme_solid"           "theme_stata"          
[21] "theme_tufte"           "theme_wsj"

Calc Theme

# Theme similar to the default settings of LibreOffice Calc
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_calc()

The economist theme

# Style plots similar to those in The Economist
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_economist()

The economist white theme

# Style plots similar to those in The Economist
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_economist_white()

Few theme

# Theme based on the rules and examples in Stephen Few, 'Practical rules for
# using colors in charts'
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_few()

Solarized theme

# Check http://ethanschoonover.com/solarized for a description of the
# solarized palette
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_solarized()

gdocs theme

# Theme similar to the default used in google docs
ggplot() + geom_point(data = iris, aes(x = Petal.Width, y = Petal.Length, color = Species)) + 
    theme_gdocs()

Classification and Resampling

Lift charts

library(ISLR)
library(tidyverse)
data("Default")
library(caret)
logit_fit3 <- glm(default ~ balance, family = binomial, data = Default)
preds_DF <- data.frame(preds = predict(logit_fit3, type = "response"), true = factor(Default$default, 
    levels = c("Yes", "No")))
creditLift <- lift(true ~ preds, data = preds_DF)
xyplot(creditLift, main = "X = balance")

library("caret")
logit_fit2 <- glm(default ~ student, family = binomial, data = Default)
preds_DF <- data.frame(preds = predict(logit_fit2, type = "response"), true = factor(Default$default, 
    levels = c("Yes", "No")))
creditLift <- lift(true ~ preds, data = preds_DF)
xyplot(creditLift, main = "X = student")

Calibration charts

scores3DF <- data.frame(default = ifelse(Default$default == "Yes", 1, 0), scores = predict(logit_fit3, 
    type = "response"))
library(plyr)
calData <- ddply(scores3DF, .(cut(scores3DF$scores, c(0, 0.05, 0.15, 0.25, 0.35, 
    0.45, 0.55, 0.65, 0.75, 0.85, 0.95, 1))), colwise(mean))
calData$midpoint <- c(0.025, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.975)
colnames(calData) <- c("preds", "true", "midpoint")
calPlot <- ggplot(calData, aes(x = midpoint, y = true)) + geom_point() + ylim(0, 
    1) + geom_abline(intercept = 0, slope = 1, color = "red") + xlab("Prediction midpoint") + 
    ylab("Observed event percentage")
plot(calPlot)

Downsampling and up-sampling

library("ROSE")
data_rose_down <- ROSE(default ~ ., data = Default, N = 666, p = 1/2)

table(data_rose_down$data$default)


 No Yes 
312 354

data_rose_up <- ROSE(default ~ ., data = Default, N = 12000, p = 1/2)
table(data_rose_up$data$default)


  No  Yes 
6025 5975

# logit downsampled model
logit_down <- glm(default ~ balance, data = data_rose_down$data, family = "binomial")
summary(logit_down)


Call:
glm(formula = default ~ balance, family = "binomial", data = data_rose_down$data)

Deviance Residuals: 
     Min        1Q    Median        3Q       Max  
-2.76567  -0.28293   0.09854   0.38613   2.89963  

Coefficients:
              Estimate Std. Error z value            Pr(>|z|)    
(Intercept) -7.1201686  0.5789729  -12.30 <0.0000000000000002 ***
balance      0.0055852  0.0004282   13.04 <0.0000000000000002 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 920.62  on 665  degrees of freedom
Residual deviance: 368.96  on 664  degrees of freedom
AIC: 372.96

Number of Fisher Scoring iterations: 6

# logit up-sampled
logit_up <- glm(default ~ balance, data = data_rose_up$data, family = "binomial")
summary(logit_up)


Call:
glm(formula = default ~ balance, family = "binomial", data = data_rose_up$data)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-3.4497  -0.3822  -0.0351   0.4427   2.9333  

Coefficients:
              Estimate Std. Error z value            Pr(>|z|)    
(Intercept) -6.6309657  0.1242331  -53.38 <0.0000000000000002 ***
balance      0.0050092  0.0000895   55.97 <0.0000000000000002 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 16635  on 11999  degrees of freedom
Residual deviance:  7577  on 11998  degrees of freedom
AIC: 7581

Number of Fisher Scoring iterations: 6

# vanilla logit
logit <- glm(default ~ balance, data = Default, family = "binomial")

# generate scores and class predictions
scores_down = predict(logit_down, type = "response")

scores_up = predict(logit_up, type = "response")

scores_reg = predict(logit, type = "response")

class_down = ifelse(scores_down > 0.5, 1, 0)
class_up = ifelse(scores_up > 0.5, 1, 0)
class_reg = ifelse(scores_reg > 0.5, 1, 0)
# > diagnostics(table(data_rose_down$data$default,class_down)) Accuracy:
# 0.872 TP: 307 TN: 274 Sensitivity: 0.9 Specificity: 0.843 False Pos Rate:
# 0.157 > diagnostics(table(data_rose_up$data$default,class_up)) Accuracy:
# 0.869 TP: 5282 TN: 5149 Sensitivity: 0.88 Specificity: 0.859 False Pos
# Rate: 0.141 > diagnostics(table(Default$default,class_reg))

Resampling Methods

# Auto dataset
library(ISLR)
library(tidyverse)
set.seed(1861)
data(Auto)

Auto_sub <- Auto %>% select(-name)
head(Auto_sub)

  mpg cylinders displacement horsepower weight acceleration year origin
1  18         8          307        130   3504         12.0   70      1
2  15         8          350        165   3693         11.5   70      1
3  18         8          318        150   3436         11.0   70      1
4  16         8          304        150   3433         12.0   70      1
5  17         8          302        140   3449         10.5   70      1
6  15         8          429        198   4341         10.0   70      1

Leave one out cross-validation (LOOCV)

# for loop of model
mods_LOOCV <- list()
preds_LOOCV <- NULL
for (i in 1:nrow(Auto)) {
    mod = lm(mpg ~ ., data = Auto_sub %>% slice(-i))
    preds_LOOCV[i] <- predict(mod, newdata = slice(Auto_sub, i))
    mods_LOOCV[[i]] <- mod
}

head(preds_LOOCV)

[1] 14.92963 13.98084 15.17965 15.04054 14.91319 10.54207

mod_insample <- lm(mpg ~ ., data = Auto_sub)

# compute RMSE LOOCV
preds_DF <- data.frame(preds_LOOCV = preds_LOOCV, preds_insample = predict(mod_insample), 
    true = Auto$mpg)

library(caret)
RMSE(preds_DF$preds_LOOCV, preds_DF$true)

[1] 3.37211

RMSE(preds_DF$preds_insample, preds_DF$true)

[1] 3.293551

R2(preds_DF$preds_LOOCV, preds_DF$true)

[1] 0.8128915

R2(preds_DF$preds_insample, preds_DF$true)

[1] 0.8214781

k-fold Cross validation

Auto_sub <- mutate(Auto_sub, folds = createFolds(Auto_sub$mpg, k = 10, list = FALSE))

### K-Fold Cross Validation
nfolds <- 10
preds_10FoldCV_DF <- data.frame(folds = Auto_sub$folds, preds_10FoldCV = rep(NA, 
    nrow(Auto_sub)))
for (i in 1:nfolds) {
    mod <- lm(mpg ~ ., data = Auto_sub %>% filter(folds != i))
    preds <- predict(mod, newdata = filter(Auto_sub, folds == i))
    preds_10FoldCV_DF[preds_10FoldCV_DF$folds == i, "preds_10FoldCV"] <- preds
}


preds_DF <- data.frame(preds_10FoldCV = preds_10FoldCV_DF$preds_10FoldCV, preds_DF)

RMSE(preds_DF$preds_10FoldCV, preds_DF$true)

[1] 3.349023

RMSE(preds_DF$preds_LOOCV, preds_DF$true)

[1] 3.37211

RMSE(preds_DF$preds_insample, preds_DF$true)

[1] 3.293551

R2(preds_DF$preds_10FoldCV, preds_DF$true)

[1] 0.8154466

R2(preds_DF$preds_LOOCV, preds_DF$true)

[1] 0.8128915

R2(preds_DF$preds_insample, preds_DF$true)

[1] 0.8214781

Bootstrapping

B = 100  # number of bootstraped datasets
n_boot = 200  # size of each bootstrapped sample
coef_boot = NULL
for (b in 1:B) {
    idx <- sample(1:nrow(Auto_sub), size = n_boot, replace = TRUE)
    mod <- lm(mpg ~ displacement, data = Auto_sub %>% slice(idx))
    coef_boot[b] <- mod$coefficients[2]
}

mod_lm <- lm(mpg ~ displacement, data = Auto_sub)

coef_boot <- data.frame(coef_boot = coef_boot)

ggplot(coef_boot, aes(x = coef_boot)) + geom_histogram() + geom_vline(xintercept = mod_lm$coefficients[2], 
    color = "red")

more complicated function for bootstrapping

library(rsample)
library(broom)
library(purrr)
library(boot)


boots <- bootstraps(Auto_sub, times = 100)
boots

# Bootstrap sampling 
# A tibble: 100 x 2
   splits            id          
   <list>            <chr>       
 1 <split [392/147]> Bootstrap001
 2 <split [392/145]> Bootstrap002
 3 <split [392/145]> Bootstrap003
 4 <split [392/148]> Bootstrap004
 5 <split [392/140]> Bootstrap005
 6 <split [392/141]> Bootstrap006
 7 <split [392/130]> Bootstrap007
 8 <split [392/140]> Bootstrap008
 9 <split [392/149]> Bootstrap009
10 <split [392/147]> Bootstrap010
# … with 90 more rows

fit_lm_on_boots <- function(split) {
    lm(mpg ~ displacement - 1, data = analysis(split))
}

boot_mods <- boots %>% mutate(model = map(splits, fit_lm_on_boots), coef_info = map(model, 
    tidy))

boot_coefs <- boot_mods %>% unnest(coef_info)

boot_coefs

# A tibble: 100 x 8
   splits      id       model term    estimate std.error statistic  p.value
   <list>      <chr>    <lis> <chr>      <dbl>     <dbl>     <dbl>    <dbl>
 1 <split [39… Bootstr… <lm>  displa…   0.0770   0.00372      20.7 8.98e-65
 2 <split [39… Bootstr… <lm>  displa…   0.0841   0.00410      20.5 6.15e-64
 3 <split [39… Bootstr… <lm>  displa…   0.0835   0.00408      20.5 7.02e-64
 4 <split [39… Bootstr… <lm>  displa…   0.0784   0.00394      19.9 2.52e-61
 5 <split [39… Bootstr… <lm>  displa…   0.0770   0.00372      20.7 1.10e-64
 6 <split [39… Bootstr… <lm>  displa…   0.0799   0.00409      19.5 9.33e-60
 7 <split [39… Bootstr… <lm>  displa…   0.0799   0.00398      20.1 4.25e-62
 8 <split [39… Bootstr… <lm>  displa…   0.0857   0.00430      19.9 2.17e-61
 9 <split [39… Bootstr… <lm>  displa…   0.0821   0.00402      20.4 1.70e-63
10 <split [39… Bootstr… <lm>  displa…   0.0815   0.00394      20.7 1.01e-64
# … with 90 more rows

ggplot(boot_coefs, aes(estimate)) + geom_histogram(bins = 20)

Forward Stepwise Selection

library("leaps")

auto_fit_fwd <- regsubsets(mpg ~ ., data = Auto_sub, nvmax = 7, method = "forward")

summary(auto_fit_fwd)

Subset selection object
Call: regsubsets.formula(mpg ~ ., data = Auto_sub, nvmax = 7, method = "forward")
8 Variables  (and intercept)
             Forced in Forced out
cylinders        FALSE      FALSE
displacement     FALSE      FALSE
horsepower       FALSE      FALSE
weight           FALSE      FALSE
acceleration     FALSE      FALSE
year             FALSE      FALSE
origin           FALSE      FALSE
folds            FALSE      FALSE
1 subsets of each size up to 7
Selection Algorithm: forward
         cylinders displacement horsepower weight acceleration year origin
1  ( 1 ) " "       " "          " "        "*"    " "          " "  " "   
2  ( 1 ) " "       " "          " "        "*"    " "          "*"  " "   
3  ( 1 ) " "       " "          " "        "*"    " "          "*"  "*"   
4  ( 1 ) " "       "*"          " "        "*"    " "          "*"  "*"   
5  ( 1 ) " "       "*"          "*"        "*"    " "          "*"  "*"   
6  ( 1 ) "*"       "*"          "*"        "*"    " "          "*"  "*"   
7  ( 1 ) "*"       "*"          "*"        "*"    " "          "*"  "*"   
         folds
1  ( 1 ) " "  
2  ( 1 ) " "  
3  ( 1 ) " "  
4  ( 1 ) " "  
5  ( 1 ) " "  
6  ( 1 ) " "  
7  ( 1 ) "*"

plot(auto_fit_fwd, scale = "adjr2")

Backward Stepwise Selection

auto_fit_bkwd <- regsubsets(mpg ~ ., data = Auto_sub, nvmax = 7, method = "backward")
summary(auto_fit_bkwd)

Subset selection object
Call: regsubsets.formula(mpg ~ ., data = Auto_sub, nvmax = 7, method = "backward")
8 Variables  (and intercept)
             Forced in Forced out
cylinders        FALSE      FALSE
displacement     FALSE      FALSE
horsepower       FALSE      FALSE
weight           FALSE      FALSE
acceleration     FALSE      FALSE
year             FALSE      FALSE
origin           FALSE      FALSE
folds            FALSE      FALSE
1 subsets of each size up to 7
Selection Algorithm: backward
         cylinders displacement horsepower weight acceleration year origin
1  ( 1 ) " "       " "          " "        "*"    " "          " "  " "   
2  ( 1 ) " "       " "          " "        "*"    " "          "*"  " "   
3  ( 1 ) " "       " "          " "        "*"    " "          "*"  "*"   
4  ( 1 ) " "       "*"          " "        "*"    " "          "*"  "*"   
5  ( 1 ) " "       "*"          "*"        "*"    " "          "*"  "*"   
6  ( 1 ) "*"       "*"          "*"        "*"    " "          "*"  "*"   
7  ( 1 ) "*"       "*"          "*"        "*"    " "          "*"  "*"   
         folds
1  ( 1 ) " "  
2  ( 1 ) " "  
3  ( 1 ) " "  
4  ( 1 ) " "  
5  ( 1 ) " "  
6  ( 1 ) " "  
7  ( 1 ) "*"

plot(auto_fit_bkwd, scale = "adjr2")

Classification and Regression

You can use str to get info about what is contained in a model ie: str(mod1) ## Setup Test/Train

train_idx <- sample(1:nrow(movies), size = floor(0.75 * nrow(movies)))
movies_train <- movies %>% slice(train_idx)
movies_test <- movies %>% slice(-train_idx)

Where 0.75 is the percentage (75%) of the data to put in the Training set.

Linear Regression

Linear regression is a linear approach to modeling the relationship between a scalar response (or dependent variable) and one or more explanatory variables (or independent variables). The case of one explanatory variable is called simple linear regression. Generate a linear model with lm(), desired formula is written with the dependant variable followed by ~ and then a list of the independant variables Can use . for all, or do something like y ~ -director Can get the coefficients like this mod1$coefficients[1]

mod1 <- lm(gross ~ budget + duration, data = movies_train)
summary(mod1)


Call:
lm(formula = gross ~ budget + duration, data = movies_train)

Residuals:
       Min         1Q     Median         3Q        Max 
-398660902  -23412577   -9492452   12647133  495725939 

Coefficients:
                   Estimate      Std. Error t value             Pr(>|t|)
(Intercept) -12355521.86555   4930418.20639  -2.506               0.0123
budget              0.99734         0.02389  41.752 < 0.0000000000000002
duration       229023.81048     45694.54514   5.012          0.000000571
               
(Intercept) *  
budget      ***
duration    ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 53860000 on 2905 degrees of freedom
  (496 observations deleted due to missingness)
Multiple R-squared:  0.4172,    Adjusted R-squared:  0.4168 
F-statistic:  1040 on 2 and 2905 DF,  p-value: < 0.00000000000000022

Logistic Regression

Logistic regression is a statistical model that in its basic form uses a logistic function to model a binary dependent variable, although many more complex extensions exist. In regression analysis, logistic regression (or logit regression) is estimating the parameters of a logistic model (a form of binary regression). Use function glm() notice the family = binomial

library(ISLR)
data(Default)
options(scipen = 9)
logitMod1 <- glm(factor(default) ~ balance, family = binomial, data = Default)

summary(logitMod1)


Call:
glm(formula = factor(default) ~ balance, family = binomial, data = Default)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-2.2697  -0.1465  -0.0589  -0.0221   3.7589  

Coefficients:
               Estimate  Std. Error z value Pr(>|z|)    
(Intercept) -10.6513306   0.3611574  -29.49   <2e-16 ***
balance       0.0054989   0.0002204   24.95   <2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 2920.6  on 9999  degrees of freedom
Residual deviance: 1596.5  on 9998  degrees of freedom
AIC: 1600.5

Number of Fisher Scoring iterations: 8

round(logitMod1$coefficients, 4)

(Intercept)     balance 
   -10.6513      0.0055

logitMod2 <- glm(factor(default) ~ balance + student + income, family = binomial, 
    data = Default)

Ridge Regression

library(glmnet)
library(glmnetUtils)

# load the movies dataset
library("tidyverse")
options(scipen = 50)
set.seed(1861)
movies <- read.csv(here::here("datasets", "movie_metadata.csv"))
movies <- movies %>% filter(budget < 400000000) %>% filter(content_rating != 
    "", content_rating != "Not Rated", !is.na(gross))
movies <- movies %>% mutate(genre_main = unlist(map(strsplit(as.character(movies$genres), 
    "\\|"), 1)), grossM = gross/1000000, budgetM = budget/1000000, profitM = grossM - 
    budgetM)
movies <- movies %>% mutate(genre_main = fct_lump(genre_main, 5), content_rating = fct_lump(content_rating, 
    3), country = fct_lump(country, 2), cast_total_facebook_likes000s = cast_total_facebook_likes/1000, 
    ) %>% drop_na()

train_idx <- sample(1:nrow(movies), size = floor(0.75 * nrow(movies)))
movies_train <- movies %>% slice(train_idx)
movies_test <- movies %>% slice(-train_idx)

# estimate ridge mod
Ridge_mod <- cv.glmnet(profitM ~ ., data = movies_train %>% select(-c(director_name, 
    actor_1_name, actor_2_name, actor_3_name, plot_keywords, movie_imdb_link, 
    country, budgetM, grossM, genres, language, movie_title)), alpha = 0)

coef(Ridge_mod)

31 x 1 sparse Matrix of class "dgCMatrix"
                                             1
(Intercept)                   353.116523189860
color                           3.220429457000
color Black and White          -1.230889234459
colorColor                      1.273946091430
num_critic_for_reviews         -0.004287598868
duration                       -0.059435354254
director_facebook_likes        -0.000095603299
actor_3_facebook_likes          0.000377629504
actor_1_facebook_likes         -0.000046854225
gross                           0.000000808363
num_voted_users                 0.000026098607
cast_total_facebook_likes       0.000011936125
facenumber_in_poster            0.097107398006
num_user_for_reviews            0.001753590606
content_ratingPG                1.541944591659
content_ratingPG-13            -0.081014605835
content_ratingR                -0.790656664619
content_ratingOther             0.262087870726
budget                         -0.000000764324
title_year                     -0.175902433003
actor_2_facebook_likes         -0.000072922047
imdb_score                      0.793676412085
aspect_ratio                   -1.260327725766
movie_facebook_likes            0.000031219836
genre_mainAction               -2.508682376558
genre_mainAdventure            -0.949174857631
genre_mainComedy                1.857795657356
genre_mainCrime                -0.752570921384
genre_mainDrama                 0.358810332201
genre_mainOther                 1.660817121433
cast_total_facebook_likes000s   0.021206225810

# explore how coefficients change as we change lambda
library(coefplot)
coefpath(Ridge_mod)

Lasso Regression

library(glmnet)
library(glmnetUtils)

# load the movies dataset
library("tidyverse")
options(scipen = 50)
set.seed(1861)
movies <- read.csv(here::here("datasets", "movie_metadata.csv"))
movies <- movies %>% filter(budget < 400000000) %>% filter(content_rating != 
    "", content_rating != "Not Rated", !is.na(gross))
movies <- movies %>% mutate(genre_main = unlist(map(strsplit(as.character(movies$genres), 
    "\\|"), 1)), grossM = gross/1000000, budgetM = budget/1000000, profitM = grossM - 
    budgetM)
movies <- movies %>% mutate(genre_main = fct_lump(genre_main, 5), content_rating = fct_lump(content_rating, 
    3), country = fct_lump(country, 2), cast_total_facebook_likes000s = cast_total_facebook_likes/1000) %>% 
    drop_na() %>% select(-c(director_name, actor_1_name, actor_2_name, actor_3_name, 
    plot_keywords, movie_imdb_link, country, budgetM, grossM, genres, language, 
    movie_title, budget, gross))

train_idx <- sample(1:nrow(movies), size = floor(0.75 * nrow(movies)))
movies_train <- movies %>% slice(train_idx)
movies_test <- movies %>% slice(-train_idx)
# estimate Lasso mod
Lasso_mod <- 
cv.glmnet(profitM ~ ., data = movies_train, alpha = 1)

coef(Lasso_mod, s = Lasso_mod$lambda.min)

29 x 1 sparse Matrix of class "dgCMatrix"
                                            1
(Intercept)                   1704.5386926912
color                            7.8685464686
color Black and White          -14.0683095963
colorColor                       .           
num_critic_for_reviews          -0.0114947113
duration                        -0.2776782023
director_facebook_likes         -0.0005768620
actor_3_facebook_likes          -0.0095188709
actor_1_facebook_likes          -0.0081385588
num_voted_users                  0.0001681242
cast_total_facebook_likes        0.0080395751
facenumber_in_poster             0.2666746699
num_user_for_reviews             0.0075827386
content_ratingPG                10.8248014912
content_ratingPG-13              .           
content_ratingR                 -7.3220267644
content_ratingOther              1.7707853864
title_year                      -0.8436939485
actor_2_facebook_likes          -0.0083079361
imdb_score                       3.1091588898
aspect_ratio                    -5.7165591209
movie_facebook_likes             0.0001270866
genre_mainAction               -11.7644203113
genre_mainAdventure             -2.7208186957
genre_mainComedy                 6.1636778997
genre_mainCrime                 -6.8162401714
genre_mainDrama                  .           
genre_mainOther                  5.3857834335
cast_total_facebook_likes000s    0.0141678626

coef(Lasso_mod, s = Lasso_mod$lambda.1se)

29 x 1 sparse Matrix of class "dgCMatrix"
                                           1
(Intercept)                   987.3257625732
color                           .           
color Black and White           .           
colorColor                      .           
num_critic_for_reviews          .           
duration                       -0.0650488589
director_facebook_likes         .           
actor_3_facebook_likes          0.0010865538
actor_1_facebook_likes          .           
num_voted_users                 0.0001630243
cast_total_facebook_likes       .           
facenumber_in_poster            .           
num_user_for_reviews            .           
content_ratingPG                5.7811040203
content_ratingPG-13             .           
content_ratingR                -4.3764504437
content_ratingOther             .           
title_year                     -0.4879797143
actor_2_facebook_likes          .           
imdb_score                      .           
aspect_ratio                   -2.7584906529
movie_facebook_likes            .           
genre_mainAction               -5.3686769625
genre_mainAdventure             .           
genre_mainComedy                3.9983665305
genre_mainCrime                 .           
genre_mainDrama                 .           
genre_mainOther                 .           
cast_total_facebook_likes000s   .

# put in a matrix
coef_mat <- data.frame(varname = rownames(coef(Lasso_mod)) %>% data.frame(), 
    Lasso_min = as.matrix(coef(Lasso_mod, s = Lasso_mod$lambda.min)) %>% round(3), 
    Lasso_1se = as.matrix(coef(Lasso_mod, s = Lasso_mod$lambda.1se)) %>% round(3)) %>% 
    rename(varname = 1, Lasso_min = 2, Lasso_1se = 3) %>% remove_rownames()

coef_mat

                     varname Lasso_min Lasso_1se
1                (Intercept)  1704.539   987.326
2                      color     7.869     0.000
3      color Black and White   -14.068     0.000
4                 colorColor     0.000     0.000
5     num_critic_for_reviews    -0.011     0.000
6                   duration    -0.278    -0.065
7    director_facebook_likes    -0.001     0.000
8     actor_3_facebook_likes    -0.010     0.001
9     actor_1_facebook_likes    -0.008     0.000
10           num_voted_users     0.000     0.000
11 cast_total_facebook_likes     0.008     0.000
12      facenumber_in_poster     0.267     0.000
13      num_user_for_reviews     0.008     0.000
14          content_ratingPG    10.825     5.781
15       content_ratingPG-13     0.000     0.000
16           content_ratingR    -7.322    -4.376
17       content_ratingOther     1.771     0.000
18                title_year    -0.844    -0.488
19    actor_2_facebook_likes    -0.008     0.000
20                imdb_score     3.109     0.000
21              aspect_ratio    -5.717    -2.758
22      movie_facebook_likes     0.000     0.000
23          genre_mainAction   -11.764    -5.369
24       genre_mainAdventure    -2.721     0.000
25          genre_mainComedy     6.164     3.998
 [ reached 'max' / getOption("max.print") -- omitted 4 rows ]

plot(Lasso_mod)

# place in one coef


# explore how coefficients change as we change lambda
library(coefplot)
coefpath(Lasso_mod)

ElasticNet

alpha_list <- seq(0, 1, len = 5)
alpha_list

[1] 0.00 0.25 0.50 0.75 1.00

enet_fit <- cva.glmnet(profitM ~ ., data = movies_train, alpha = alpha_list)

enet_fit

Call:
cva.glmnet.formula(formula = profitM ~ ., data = movies_train, 
    alpha = alpha_list)

Model fitting options:
    Sparse model matrix: FALSE
    Use model.frame: FALSE
    Alpha values: 0 0.25 0.5 0.75 1
    Number of crossvalidation folds for lambda: 10

minlossplot(enet_fit)

plot(enet_fit)

# look at each individual model
plot(enet_fit$modlist[[4]])

# view coefficients
coef(enet_fit, alpha = 0.75) %>% round(3)

29 x 1 sparse Matrix of class "dgCMatrix"
                                    1
(Intercept)                   967.491
color                           .    
color Black and White           .    
colorColor                      .    
num_critic_for_reviews          .    
duration                       -0.057
director_facebook_likes         .    
actor_3_facebook_likes          0.001
actor_1_facebook_likes          .    
num_voted_users                 0.000
cast_total_facebook_likes       .    
facenumber_in_poster            .    
num_user_for_reviews            .    
content_ratingPG                5.704
content_ratingPG-13             .    
content_ratingR                -4.308
content_ratingOther             .    
title_year                     -0.478
actor_2_facebook_likes          .    
imdb_score                      .    
aspect_ratio                   -2.778
movie_facebook_likes            .    
genre_mainAction               -5.158
genre_mainAdventure             .    
genre_mainComedy                3.928
genre_mainCrime                 .    
genre_mainDrama                 .    
genre_mainOther                 .    
cast_total_facebook_likes000s   .

# if we want the lambda.min version
coef(enet_fit, alpha = 0.75, s = enet_fit$modlist[[4]]$lambda.min)

29 x 1 sparse Matrix of class "dgCMatrix"
                                            1
(Intercept)                   1704.8100887902
color                            8.3117548645
color Black and White          -14.1284756012
colorColor                       .           
num_critic_for_reviews          -0.0119109398
duration                        -0.2792141763
director_facebook_likes         -0.0005735128
actor_3_facebook_likes          -0.0098836808
actor_1_facebook_likes          -0.0083783098
num_voted_users                  0.0001680881
cast_total_facebook_likes        0.0056239902
facenumber_in_poster             0.2695339896
num_user_for_reviews             0.0076388802
content_ratingPG                10.8481088871
content_ratingPG-13              .           
content_ratingR                 -7.3148179137
content_ratingOther              1.8032278465
title_year                      -0.8438615745
actor_2_facebook_likes          -0.0085528071
imdb_score                       3.1407490708
aspect_ratio                    -5.7212843610
movie_facebook_likes             0.0001285754
genre_mainAction               -11.8516721784
genre_mainAdventure             -2.8045684545
genre_mainComedy                 6.0821880425
genre_mainCrime                 -6.8999275318
genre_mainDrama                  .           
genre_mainOther                  5.3565127462
cast_total_facebook_likes000s    2.6695073318

coef(enet_fit, alpha = 1) %>% round(3)

29 x 1 sparse Matrix of class "dgCMatrix"
                                    1
(Intercept)                   987.326
color                           .    
color Black and White           .    
colorColor                      .    
num_critic_for_reviews          .    
duration                       -0.065
director_facebook_likes         .    
actor_3_facebook_likes          0.001
actor_1_facebook_likes          .    
num_voted_users                 0.000
cast_total_facebook_likes       .    
facenumber_in_poster            .    
num_user_for_reviews            .    
content_ratingPG                5.781
content_ratingPG-13             .    
content_ratingR                -4.376
content_ratingOther             .    
title_year                     -0.488
actor_2_facebook_likes          .    
imdb_score                      .    
aspect_ratio                   -2.758
movie_facebook_likes            .    
genre_mainAction               -5.369
genre_mainAdventure             .    
genre_mainComedy                3.998
genre_mainCrime                 .    
genre_mainDrama                 .    
genre_mainOther                 .    
cast_total_facebook_likes000s   .

enet_coefs <- data.frame(varname = rownames(coef(enet_fit, alpha = 0)), ridge = as.matrix(coef(enet_fit, 
    alpha = 0)) %>% round(3), alpha025 = as.matrix(coef(enet_fit, alpha = 0.25)) %>% 
    round(3), alpha05 = as.matrix(coef(enet_fit, alpha = 0.5)) %>% round(3), 
    alpha075 = as.matrix(coef(enet_fit, alpha = 0.75)) %>% round(3), lasso = as.matrix(coef(enet_fit, 
        alpha = 1)) %>% round(3)) %>% rename(varname = 1, ridge = 2, alpha025 = 3, 
    alpha05 = 4, alpha075 = 5, lasso = 6) %>% remove_rownames()

Predict

scores <- predict(logitMod1, type = "response")
head(scores)

           1            2            3            4            5 
0.0013056797 0.0021125949 0.0085947405 0.0004344368 0.0017769574 
           6 
0.0037041528

preds_DF <- data.frame(scores_mod1 = predict(logitMod1, type = "response"), 
    scores_mod2 = predict(logitMod2, type = "response"), Default)

Confusion Matrix

library("caret")
confusionMatrix(factor(ifelse(preds_DF$scores_mod1 > 0.5, "Yes", "No"), levels = c("Yes", 
    "No")), factor(preds_DF$default, levels = c("Yes", "No")))

Confusion Matrix and Statistics

          Reference
Prediction  Yes   No
       Yes  100   42
       No   233 9625
                                               
               Accuracy : 0.9725               
                 95% CI : (0.9691, 0.9756)     
    No Information Rate : 0.9667               
    P-Value [Acc > NIR] : 0.0004973            
                                               
                  Kappa : 0.4093               
                                               
 Mcnemar's Test P-Value : < 0.00000000000000022
                                               
            Sensitivity : 0.3003               
            Specificity : 0.9957               
         Pos Pred Value : 0.7042               
         Neg Pred Value : 0.9764               
             Prevalence : 0.0333               
         Detection Rate : 0.0100               
   Detection Prevalence : 0.0142               
      Balanced Accuracy : 0.6480               
                                               
       'Positive' Class : Yes

ROC Curve

TrainDF <- data.frame(default = c(Default$default, Default$default), scores = c(preds_DF$scores_mod1, 
    preds_DF$scores_mod2), models = c(rep("X = Student", length(preds_DF$scores_mod1)), 
    rep("X = Student + Balance + Income", length(preds_DF$scores_mod2))))


library(ggplot2)
library("plotROC")
TrainROC <- ggplot(TrainDF, aes(m = scores, d = default, color = models)) + 
    geom_roc(show.legend = TRUE, labelsize = 3.5, cutoffs.at = c(0.99, 0.9, 
        0.7, 0.5, 0.3, 0.1, 0))
TrainROC <- TrainROC + style_roc(theme = theme_grey) + theme(axis.text = element_text(colour = "blue")) + 
    theme(legend.justification = c(1, 0), legend.position = c(1, 0), legend.box.margin = margin(c(50, 
        50, 50, 50)))
plot(TrainROC)

Calibration Plot

load library(plyr) before library(tidyverse)

scores3DF <- data.frame(default = ifelse(Default$default == "Yes", 1, 0), scores = preds_DF$scores_mod2)

calData <- ddply(scores3DF, .(cut(scores3DF$scores, c(0, 0.05, 0.15, 0.25, 0.35, 
    0.45, 0.55, 0.65, 0.75, 0.85, 0.95, 1))), colwise(mean))
calData$midpoint <- c(0.025, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.975)
colnames(calData) <- c("preds", "true", "midpoint")
calPlot <- ggplot(calData, aes(x = midpoint, y = true)) + geom_point() + ylim(0, 
    1) + geom_abline(intercept = 0, slope = 1, color = "red") + xlab("Prediction midpoint") + 
    ylab("Observed event percentage")

plot(calPlot)

ForCats (working with factors)

R uses factors to handle categorical variables, variables that have a fixed and known set of possible values. Factors are also helpful for reordering character vectors to improve display. The goal of the forcats package is to provide a suite of tools that solve common problems with factors, including changing the order of levels or the values. Some examples include:

fct_reorder(): Reordering a factor by another variable.
fct_infreq(): Reordering a factor by the frequency of values.
fct_relevel(): Changing the order of a factor by hand.
fct_lump(): Collapsing the least/most frequent values of a factor into “other”.

Exam Reference - Material Theme

2019-10-28

Themes for this doc

Basics

Reading Data

Setup Test/Train

Containers

Factors

dplyr (Data Manipulation)

Filter

Arange

slice

select

Select Helpers

rename

mutate

group_by

Summerize

Pulling it together

magrittr (pipe opperator)

Additional Pipe Opperators

Tee %T>%

Exposition %$%

Compound assignment %<>%

Plot

ggplot

Scatter

Bar Plot

Jitter

Histogram

Box Plot

Line

Error Bars

Themes

ggthemes

Calc Theme

The economist theme

The economist white theme

Few theme

Solarized theme

gdocs theme

Classification and Resampling

Lift charts

Calibration charts

Downsampling and up-sampling

Resampling Methods

Leave one out cross-validation (LOOCV)

k-fold Cross validation

Bootstrapping

more complicated function for bootstrapping

Forward Stepwise Selection

Backward Stepwise Selection

Classification and Regression

Linear Regression

Logistic Regression

Ridge Regression

Lasso Regression

ElasticNet

Predict

Confusion Matrix

ROC Curve

Calibration Plot

ForCats (working with factors)